Brain-to-brain coupling forecasts future joint action outcomes

Summary In this study, we investigated whether brain-to-brain coupling patterns could predict performance in a time-estimation task that requires two players to cooperate. The participant pairs were tasked with synchronizing button presses after converging on a shared representation of “short,” “medium,” and “long” time intervals while utilizing feedback to adjust responses. We employed electroencephalogram (EEG)-hyperscanning and focused on post-feedback brain activity. We found that negative feedback led to increased frontal mid-line theta activity across individuals. Moreover, a correlation in post-feedback theta power between players forecasted failed joint action, while an anti-correlation forecasted success. These findings suggest that temporally coupled feedback-related brain activity between two individuals serves as an indicator of redundancy in adjustment of a common goal representation. Additionally, the anti-correlation of this activity reflects cognitive strategic mechanisms that ensure optimal joint action outcomes. Rather than a paired overcompensation, successful cooperation requires flexible strategic agility from both partners.


Figure S1
. Response times of player 1 (blue) and player 2 (orange) for pair one as an example in function of trial number (note that the trial numbers are dummy coded due to the experimental design, wherein the conditions were interleaved and randomised for each pair, and thus these represent the duration of the experiment rather than the real trial number) for each condition (top row: low, middle row: medium, and bottom row: high).A comprehensive set of similar figures (for other pairs) can be found on: https://osf.io/ct8jb/.We also provide the code used for plotting along with dataset itself, enabling users to adjust the axis limits as needed, particularly in case of outlying responses.Related to Figure 2 and Table 1 and Table 2. Figure S2.The average frequency of trials where successful cooperation is followed by another successful cooperation (light grey) compared to trials where unsuccessful cooperation is followed by successful cooperation (dark grey) across each condition type (long, medium, and short).Individual data points are represented by black dots, and error bars indicate the standard error.We used a 2 (trial type: Correct followed by correct vs. incorrect followed by correct) x 3 (condition: short, medium, and long) repeated measures ANOVA to compare the frequency of trials where a successful cooperation was followed by another successful trial against those where an unsuccessful cooperation was followed by a successful one for each condition type (short, medium, and long).Overall, we found that participants were more likely to successfully cooperate if the previous trial (of the same type) was also successful.We found that there was a significant main effect of trial type, F(1, 28) = 40.23,p < .001,revealing that there was a greater frequency of correct followed by correct trials (M=37.5,SE=2.47) compared to incorrect followed by correct trials (M=18.6,SE=0.565).There was also a significant main effect of condition F(1.65, 46.20) = 76.26,p < .001,with a post hoc one-way ANOVA showing that pairs were more likely to successfully cooperate following a previous successful trial compared to when the previous trial was unsuccessful in the short (M difference = 43.93,F(1, 28) = 99.90, p < .001)and medium (M difference = 9.76, F(1, 28) = 7.64, p = .010)conditions, but this was not observed in the long condition (M difference = 2.93, F(1, 28) = 1.05, p = .314).Further, the 2 x 3 repeated measures ANOVA revealed a significant interaction between trial type and condition, F(1.58, 44.23) = 71.91,p < .001.The results from pairwise comparisons are displayed in Table S1.Related to Figure 2.There was a marginal effect when comparing the long and short conditions, with the first instance of successful cooperation tending to occur earlier in the short condition compared to the long condition (p = .059).No significant difference was found between the long and medium conditions (p = .985).significant condition differences (p < 0.05, cluster corrected).Firstly, we observed significant condition differences (p = .003and p <.001 respectively) in theta and delta activity maximal over the central and centro-parietal channels.Specifically, theta (and delta) power was stronger right after the onset of the feedback text in the incorrect condition until around 0.4sec compared to the correct condition.It has been previously proposed that mid-frontal theta activity serves as a neural indicator of monitoring ongoing actions and signalling unfavourable action outcomes.This in turn results in adaptive response and modulated activity due to committed errors 1,2 .Further, there was a significant condition difference (p < .001),with greater delta power in the incorrect vs. correct condition.This effect corresponded to a cluster maximal over the central electrodes that stretched between 1.15 to 1.75sec post feedback text onset.We consider this effect to be a representation of the sustained ERP in the similar time window.Finally for the delta band, we found significantly more pronounced activity in the incorrect compared to the correct condition following the feedback onset (p = .006).

Related to
This effect was most prominent between 2.65sec and 2.75sec post feedback text onset and maximal over right centro-temporal channels.Further, in the theta band, there was a more sustained theta attenuation in the incorrect compared to the correct condition (p < .001).This effect was maximal over centro-parietal and parietal sites and most prominent between 0.6 to 1.15sec post feedback text onset.We take this as spectral leakage from the effect seen in the alpha band, see below for the interpretation of this effect.Lastly for the theta band, we observed a condition effect that was maximal over temporal electrodes and left lateralised (p = .039).Specifically, there was a reduction in theta power in the incorrect condition relative to the correct condition corresponding to a cluster that stretched between 1.5 and 1.6sec post feedback onset.We also found significant condition differences in alpha activity.Firstly, there was a greater alpha power decrease in the incorrect condition relative to the correct condition (p < .001).This effect corresponded to a cluster that extended between 0.3 to 1.75sec post feedback and was maximal over central electrodes.We interpret this post error alpha decrease as stronger engagement, greater resource allocation and active processing to prevent future errors.This enhanced and more alert cognitive state optimises visual processing and better motor system coordination for future trials 1 .We also found that there was a greater alpha power increase in the correct condition relative to the incorrect condition (p = .031).This effect corresponded to a cluster that spread between 2 to 2.25sec and was maximal over left temporal channels.
We found a significant condition effect in the low beta band (p =.024).This corresponded to a transient time window of .2 to .25secpost feedback and was maximal over centro-parietal electrodes.
Further, there was a prolonged opposite pattern of low beta (p < .001)and high beta (p < .001)activity between the conditions.The incorrect condition yielded a prolonged low beta decrease, whereas the correct condition showed a prolonged low beta increase.This effect corresponded to a cluster that spanned from .35 to 1.5sec and was prominent all over the scalp.The beta power increase following correct responses and thus feedback is likely reflecting the requirement for strategy maintenance by facilitating the reinforcement of the current motor and cognitive strategy [3][4][5][6][7] .Lastly, we found a transient condition effect (p = .013)in the high beta band that extended between 2.9 and 3sec post feedback.Specifically, there was a greater high beta power increase in the incorrect compared to the correct condition, maximal over bilateral temporal sites.
(B) Feedback locked averaged ERPs produced by correct (blue) and incorrect (red) trials in the timing task averaged over all individuals (N = 59).The ERP waveforms show averaged ERPs across the overlapping (across the three time windows of the significant between condition effects) electrode clusters (Cz, CP1, Pz, and CPz) that indicate the maximal condition difference (a schematic view of these electrodes is shown in the top right corner).The shaded areas around the ERP waves represent standard error.The grey rectangles represent time windows of the significant between condition differences.The black dots in the headplots illustrate the clusters of electrodes showing most pronounced mean condition differences for each time window.The cluster-based permutation tests revealed that there was a significant effect of condition, specifically the incorrect condition was associated with a greater positivity compared to the correct condition (p < .001).This effect corresponded to a cluster that spanned from .178 to .926sec and was maximal over centro-parietal electrodes.Secondly, we found another condition effect, where the incorrect trials yielded a sustained negativity ERP relative to the correct trials (p < .001,p = .034).This corresponded to clusters that extended from 1.406 to 2.554sec as well as 2.562 to 2.634sec respectively, most prominent over centro-parietal channels.Previous research has proposed that the localization of error-related negativity (ERN) and theta activity overlap, suggesting a shared involvement in error monitoring processes 1 .Additionally, it has been suggested that the ERN serves not only as an error detection mechanism but also predicts the extent of learning from errors 8 .Consequently, we interpret this ERP effect playing a role in error monitoring and adjustment for future actions.Table S1.Post-hoc multiple comparisons (with Bonferroni correction) for the frequency of trials where a successful cooperation was followed by another successful trial (Correct followed by Correct) against those where an unsuccessful cooperation was followed by a successful trial (Incorrect followed by Correct) between conditions (short, medium, and long).Adjusted p indicates the p-value after Bonferroni correction.Related to Figure 2.

Figure S3 .
Figure S3.The average trial number at which the first successful cooperation occurred for each condition type (long, medium, and short).Individual data points are represented by black dots, and error bars indicate the standard error.We examined the differences in the trial numbers of the first successful cooperation among the different conditions (long, medium, and short) using a one-way ANOVA.We found a significant effect of condition on the trial number at which the first successful cooperation occurred (F(2, 84) = 3.85, p = 0.025).Post-hoc Tukey tests indicated a significant difference between the medium and short conditions, showing that the first instance of successful cooperation occurred earlier in the short condition compared to the medium condition (p = .039).

Figure S4 .
Figure S4.TFR of power and ERPs condition comparisons (incorrect vs. correct) locked to the onset of feedback.Related to Figure 3. (A) TFRs of power (at the Cz electrode) for incorrect trials (left), correct trials (middle), and the condition differences (i.e., incorrectcorrect) (right) averaged over all individuals (N = 59), locked to the onset of feedback text.Headplots illustrate the clusters of electrodes that show the most pronounced mean condition difference.Black rectangles indicate

Figure S5 .
Figure S5.Forecasting successful (blue) and unsuccessful (orange) joint action outcomes on next consecutive trials based on the correlation values (Fisher Z) of oscillatory power coupling between player 1 and player 2 after the onset of feedback text in the failed cooperation condition.Line plots are representing the correlation (Fisher Z) values averaged over the electrodes from all over the scalp (the averaged electrodes are those that show maximal condition differences between correct and incorrect trials in each frequency band; however, note that the electrodes are different for each frequency band; see Fig. S4A) in the alpha (8-14Hz; left panel), low beta (15-20Hz; middle panel), and high beta (20-25Hz; right panel) frequency bands.Note that there are no significant condition differences in either of the frequency bands (p < .0.05, cluster corrected).Related to Figure 4.